- 阅读权限
- 255
- 威望
- 3 级
- 论坛币
- 393509 个
- 通用积分
- 180.6551
- 学术水平
- 715 点
- 热心指数
- 890 点
- 信用等级
- 660 点
- 经验
- 139762 点
- 帖子
- 13600
- 精华
- 3
- 在线时间
- 4701 小时
- 注册时间
- 2009-7-17
- 最后登录
- 2023-1-16
|
- package org.scalaml.supervised.bayes
- // Scala classes
- import scala.language.implicitConversions
- import scala.util.Try
- import org.apache.log4j.Logger
- import org.scalaml.core.Types.ScalaMl._
- import org.scalaml.stats.{XTSeries, Stats}
- import org.scalaml.validation.MultiFValidation
- import org.scalaml.core.ITransform
- import org.scalaml.supervised.Supervised
- import org.scalaml.util.LoggingUtils._
- import NaiveBayesModel._, XTSeries._, Stats._, NaiveBayes._
-
- /**
- * Generic Binomial Naive Bayes classification class. The class is used for both training
- * and run-time classification. The training of the model is executed during the instantiation
- * of the class to avoid having an uninitialized model. A conversion from a parameterized
- * array, ''Array[T]'' to an array of double, ''DblArray'', has to be implicitly defined for
- * training the model.
- *
- * The implemantation follows the standard design of supervised learning algorithm:
- * - The classifier implements the '''ITransform''' implicit monadic data transformation
- * - The constructor triggers the training of the classifier, making the model immutable
- * - The classifier implements the '''Monitor''' interface to collect profile information for
- * debugging purpose
- *
- * As a classifier, the method implement a data transformation '''ITransform''' with a model
- * explicitly extracted from a training set
- * {{{
- * Naive Bayes formula:
- * p(C}x) = p(x|C).p(C)/p(x) => p(C|x) = p(x1|C).p(x2|C). .. p(xn|C).p(C)
- * }}}
- * @tparam T type of features in each observation
- * @constructor Instantiate a parameterized NaiveBayes model
- * @param smoothing Laplace or Lidstone smoothing factor
- * @param xt Input labeled time series used for training
- * @param expected in the training of a model
- * @param classes Number of classes used in the Naive Bayes
- *
- * @throws IllegalArgumentException if one of the class parameters is undefined
- * @author Patrick Nicolas
- * @since 0.98 February 13, 2014
- * @version 0.99.1
- * @see Scala for Machine learning Chapter 5 "Naive Bayes Models" / Naive Bayes classifiers
- * @see org.scalaml.core.ITransform
- */
- @throws(classOf[IllegalArgumentException])
- private[scalaml] final class NaiveBayes[T <: AnyVal](
- smoothing: Double,
- xt: XVSeries[T],
- expected: Vector[Int],
- logDensity: LogDensity,
- classes: Int)(implicit f: T => Double)
- extends ITransform[Array[T]](xt) with Supervised[T, Array[T]] with Monitor[T] {
- type V = Int
- check(smoothing, xt, expected, classes)
-
- protected val logger = Logger.getLogger("NaiveBayes")
-
- /**
- * The model is instantiated during training for both
- * classes if the training is successful. It is None otherwise
- */
- val model: Option[NaiveBayesModel[T]] = Try {
- if( classes == 2)
- BinNaiveBayesModel[T](train(1), train(0))
- else
- MultiNaiveBayesModel[T](List.tabulate(classes)( train(_)))
- }
- ._toOption("NaiveBayes.train", logger)
- /**
- * Run-time classification of a time series using the Naive Bayes model. The method invoke
- * the actual classification method in one of the NaiveBayes models.
- * @throws MatchError if the input time series is undefined or have no elements or the
- * model was not properly trained
- * @return PartialFunction of time series of elements of type T as input to the Naive Bayes
- * and array of class indices as output
- */
- override def |> : PartialFunction[Array[T], Try[V]] = {
- case x: Array[T] if x.length > 0 && model.isDefined =>
- Try( model.map(_.classify(x, logDensity)).get )
- }
-
- /**
- * Compute the F1 statistics for the Naive Bayes.
- * @param xt Time series of features of type Array[T], and class indices as labels
- * @param expected expected value or label
- * @return F1 measure if the model has been properly trained, Failure otherwise
- */
- override def validate(xt: XVSeries[T], expected: Vector[V]): Try[Double] = Try {
- val predict = model.get.classify(_ : Array[T], logDensity)
- MultiFValidation[T](expected, xt, classes)(predict).score
- }
-
- /**
- * Compute the F1 statistics for the Naive Bayes.
- * @param labeled Time series of features and expected values
- * @return F1 measure if the model has been properly trained, Failure otherwise
- */
- def validate(labeled: Vector[(Array[T], V)]): Try[Double] = Try {
- val predict = model.get.classify(_: Array[T], logDensity)
- MultiFValidation[T](labeled, classes)(predict).score
- }
- /**
- * Textual representation of the Naive Bayes classifier with labels for features.
- * It returns "No Naive Bayes model" if no model exists
- * @return Stringized features with their label if model exists.
- */
- def toString(labels: Array[String]): String =
- if( model.isDefined) "No model"
- else
- if(labels.length > 0) model.get.toString(labels) else model.get.toString
-
- /**
- * Default textual representation of the Naive Bayes classifier with labels for features.
- * It returns "No Naive Bayes model" if no model exists
- * @return Stringized features with their label if model exists.
- */
- override def toString: String = toString(Array.empty[String])
-
- /**
- * Train the Naive Bayes model on one of the the classes
- */
- @throws(classOf[IllegalStateException])
- private def train(index: Int): Likelihood[T] = {
- val xv: XVSeries[Double] = xt
-
- // Extract then filter each observation to be associated to a specific label.
- // The implicit conversion from Array of type T to Array of type Double is invoked
- val values = xv.zip(expected).filter( _._2 == index).map(_._1)
- if( values.isEmpty )
- throw new IllegalStateException("NaiveBayes.train Filtered value is undefined")
-
- // Gets the dimension of a feature
- val dim = dimension(xv)
-
- // Create a likelihood instance for this class 'label'. The
- // tuple (mean, standard deviation) (2nd argument) is computed
- // by invoking XTSeries.statistics then the Lidstone mean adjustment.
- // The last argument, class likelihood p(C) is computed as the ratio of the
- // number of observations associated to this class/label over total number of observations.
- Likelihood(index,
- statistics(values).map(stat => (stat.lidstoneMean(smoothing, dim), stat.stdDev) ),
- values.size.toDouble/xv.size)
- }
- }
- /**
- * Singleton that define the constructors for the NaiveBayes classifier and
- * validate its parameters
- * @author Patrick Nicolas
- * @since 0.98 February 13, 2014
- * @version 0.98
- * @see Scala for Machine learning Chapter 5 "Naive Bayes Model"
- */
- object NaiveBayes {
-
- /**
- * Implicit conversion from a NaiveBayes[T} to a Try[NaiveBayes[T}} type.
- */
- implicit def naiveBayes2Try[T <: AnyVal](nb: NaiveBayes[T])(implicit f: T => Double): Try[NaiveBayes[T]] = Try(nb)
-
- /**
- * Default constructor for the NaiveBayes class
- * @tparam T type of features in the time series xt
- * @param smoothing Laplace or Lidstone smoothing factor
- * @param xt Input time series of observations used for training
- * @param expected Input labeled time series used for training
- * @param logDensity logarithm formulation of the probability density function
- * @param classes Number of classes used in the Naive Bayes model
- */
- def apply[T <: AnyVal](
- smoothing: Double,
- xt: XVSeries[T],
- expected: Vector[Int],
- logDensity: LogDensity,
- classes: Int)(implicit f: T => Double): NaiveBayes[T] =
- new NaiveBayes[T](smoothing, xt, expected, logDensity, classes)
-
- /**
- * Constructor for the NaiveBayes class using single input data set {observations, expected}
- * @tparam T type of features in the time series xt
- * @param smoothing Laplace or Lidstone smoothing factor
- * @param xty Input time series of pair (observations, expected outcome) used for training
- * @param logDensity logarithm formulation of the probability density function
- * @param classes Number of classes used in the Naive Bayes model
- */
- def apply[T <: AnyVal] (
- smoothing: Double,
- xty: Vector[(Array[T], Int)],
- logDensity: LogDensity = logGauss,
- classes: Int = 2)(implicit f: T => Double): NaiveBayes[T] = {
-
- val xy: (Vector[Array[T]], Vector[Int]) = xty.unzip
- new NaiveBayes[T](smoothing, xy._1, xy._2, logDensity, classes)
- }
-
- /**
- * Constructor for the Binomial NaiveBayes class using the Laplace smoothing factor, and the
- * gaussian distribution
- * @tparam T type of features in the time series xt
- * @param xt Input time series of observations used for training
- * @param expected Input labeled time series used for training
- */
- def apply[T <: AnyVal](xt: XVSeries[T], expected: Vector[Int])(implicit f: T => Double): NaiveBayes[T] =
- new NaiveBayes[T](1.0, xt, expected, logGauss, 2)
-
-
- private def check[T](
- smoothing: Double,
- xt: XVSeries[T],
- expected: Vector[Int],
- classes: Int): Unit = {
-
- require(smoothing > 0.0 && smoothing <= 1.0,
- s"NaiveBayes: Found smoothing $smoothing required 0 < smoothing <= 1")
- require( xt.nonEmpty,
- "NaiveBayes: Time series input for training Naive Bayes is undefined")
- require( expected.nonEmpty,
- "NaiveBayes: labeled values for training Naive Bayes is undefined")
- require( classes > 1,
- s"NaiveBayes: Naive Bayes found $classes required classes > 1")
- require( xt.size == expected.size,
- s"NaiveBayes: observations set ${xt.size} and labels set ${expected.size} should have same size")
- }
- }
|
-
总评分: 论坛币 + 20
查看全部评分
|
发帖
雷达卡
Scala for Machine Learning-Packt Publishing(2014).pdf
(9.4 MB, 需要: 6 个论坛币)
加JingGuanBbs
京公网安备 11010802022788号
论坛法律顾问：王进律师
知识产权保护声明
免责及隐私声明
提升卡
置顶卡
沉默卡
变色卡
抢沙发
千斤顶
显身卡
